2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
53 [N_POSSIBLE
] = NODE_MASK_ALL
,
54 [N_ONLINE
] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
60 [N_CPU
] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states
);
65 unsigned long totalram_pages __read_mostly
;
66 unsigned long totalreserve_pages __read_mostly
;
68 int percpu_pagelist_fraction
;
70 static void __free_pages_ok(struct page
*page
, unsigned int order
);
73 * results with 256, 32 in the lowmem_reserve sysctl:
74 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
75 * 1G machine -> (16M dma, 784M normal, 224M high)
76 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
77 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
78 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
80 * TBD: should special case ZONE_DMA32 machines here - in those we normally
81 * don't need any ZONE_NORMAL reservation
83 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
84 #ifdef CONFIG_ZONE_DMA
87 #ifdef CONFIG_ZONE_DMA32
96 EXPORT_SYMBOL(totalram_pages
);
98 static char * const zone_names
[MAX_NR_ZONES
] = {
99 #ifdef CONFIG_ZONE_DMA
102 #ifdef CONFIG_ZONE_DMA32
106 #ifdef CONFIG_HIGHMEM
112 int min_free_kbytes
= 1024;
114 unsigned long __meminitdata nr_kernel_pages
;
115 unsigned long __meminitdata nr_all_pages
;
116 static unsigned long __meminitdata dma_reserve
;
118 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
120 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
121 * ranges of memory (RAM) that may be registered with add_active_range().
122 * Ranges passed to add_active_range() will be merged if possible
123 * so the number of times add_active_range() can be called is
124 * related to the number of nodes and the number of holes
126 #ifdef CONFIG_MAX_ACTIVE_REGIONS
127 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
128 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
130 #if MAX_NUMNODES >= 32
131 /* If there can be many nodes, allow up to 50 holes per node */
132 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
134 /* By default, allow up to 256 distinct regions */
135 #define MAX_ACTIVE_REGIONS 256
139 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
140 static int __meminitdata nr_nodemap_entries
;
141 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
142 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
143 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
144 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
145 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
146 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
147 unsigned long __initdata required_kernelcore
;
148 unsigned long __initdata required_movablecore
;
149 unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
151 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
153 EXPORT_SYMBOL(movable_zone
);
154 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
157 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
158 EXPORT_SYMBOL(nr_node_ids
);
161 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
162 static inline int get_pageblock_migratetype(struct page
*page
)
164 return get_pageblock_flags_group(page
, PB_migrate
, PB_migrate_end
);
167 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
169 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
170 PB_migrate
, PB_migrate_end
);
173 static inline int gfpflags_to_migratetype(gfp_t gfp_flags
)
175 return ((gfp_flags
& __GFP_MOVABLE
) != 0);
179 static inline int get_pageblock_migratetype(struct page
*page
)
181 return MIGRATE_UNMOVABLE
;
184 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
188 static inline int gfpflags_to_migratetype(gfp_t gfp_flags
)
190 return MIGRATE_UNMOVABLE
;
192 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
194 #ifdef CONFIG_DEBUG_VM
195 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
199 unsigned long pfn
= page_to_pfn(page
);
202 seq
= zone_span_seqbegin(zone
);
203 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
205 else if (pfn
< zone
->zone_start_pfn
)
207 } while (zone_span_seqretry(zone
, seq
));
212 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
214 if (!pfn_valid_within(page_to_pfn(page
)))
216 if (zone
!= page_zone(page
))
222 * Temporary debugging check for pages not lying within a given zone.
224 static int bad_range(struct zone
*zone
, struct page
*page
)
226 if (page_outside_zone_boundaries(zone
, page
))
228 if (!page_is_consistent(zone
, page
))
234 static inline int bad_range(struct zone
*zone
, struct page
*page
)
240 static void bad_page(struct page
*page
)
242 printk(KERN_EMERG
"Bad page state in process '%s'\n"
243 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
244 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
245 KERN_EMERG
"Backtrace:\n",
246 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
247 (unsigned long)page
->flags
, page
->mapping
,
248 page_mapcount(page
), page_count(page
));
250 page
->flags
&= ~(1 << PG_lru
|
260 set_page_count(page
, 0);
261 reset_page_mapcount(page
);
262 page
->mapping
= NULL
;
263 add_taint(TAINT_BAD_PAGE
);
267 * Higher-order pages are called "compound pages". They are structured thusly:
269 * The first PAGE_SIZE page is called the "head page".
271 * The remaining PAGE_SIZE pages are called "tail pages".
273 * All pages have PG_compound set. All pages have their ->private pointing at
274 * the head page (even the head page has this).
276 * The first tail page's ->lru.next holds the address of the compound page's
277 * put_page() function. Its ->lru.prev holds the order of allocation.
278 * This usage means that zero-order pages may not be compound.
281 static void free_compound_page(struct page
*page
)
283 __free_pages_ok(page
, compound_order(page
));
286 static void prep_compound_page(struct page
*page
, unsigned long order
)
289 int nr_pages
= 1 << order
;
291 set_compound_page_dtor(page
, free_compound_page
);
292 set_compound_order(page
, order
);
294 for (i
= 1; i
< nr_pages
; i
++) {
295 struct page
*p
= page
+ i
;
298 p
->first_page
= page
;
302 static void destroy_compound_page(struct page
*page
, unsigned long order
)
305 int nr_pages
= 1 << order
;
307 if (unlikely(compound_order(page
) != order
))
310 if (unlikely(!PageHead(page
)))
312 __ClearPageHead(page
);
313 for (i
= 1; i
< nr_pages
; i
++) {
314 struct page
*p
= page
+ i
;
316 if (unlikely(!PageTail(p
) |
317 (p
->first_page
!= page
)))
323 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
327 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
329 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
330 * and __GFP_HIGHMEM from hard or soft interrupt context.
332 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
333 for (i
= 0; i
< (1 << order
); i
++)
334 clear_highpage(page
+ i
);
338 * function for dealing with page's order in buddy system.
339 * zone->lock is already acquired when we use these.
340 * So, we don't need atomic page->flags operations here.
342 static inline unsigned long page_order(struct page
*page
)
344 return page_private(page
);
347 static inline void set_page_order(struct page
*page
, int order
)
349 set_page_private(page
, order
);
350 __SetPageBuddy(page
);
353 static inline void rmv_page_order(struct page
*page
)
355 __ClearPageBuddy(page
);
356 set_page_private(page
, 0);
360 * Locate the struct page for both the matching buddy in our
361 * pair (buddy1) and the combined O(n+1) page they form (page).
363 * 1) Any buddy B1 will have an order O twin B2 which satisfies
364 * the following equation:
366 * For example, if the starting buddy (buddy2) is #8 its order
368 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
370 * 2) Any buddy B will have an order O+1 parent P which
371 * satisfies the following equation:
374 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
376 static inline struct page
*
377 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
379 unsigned long buddy_idx
= page_idx
^ (1 << order
);
381 return page
+ (buddy_idx
- page_idx
);
384 static inline unsigned long
385 __find_combined_index(unsigned long page_idx
, unsigned int order
)
387 return (page_idx
& ~(1 << order
));
391 * This function checks whether a page is free && is the buddy
392 * we can do coalesce a page and its buddy if
393 * (a) the buddy is not in a hole &&
394 * (b) the buddy is in the buddy system &&
395 * (c) a page and its buddy have the same order &&
396 * (d) a page and its buddy are in the same zone.
398 * For recording whether a page is in the buddy system, we use PG_buddy.
399 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
401 * For recording page's order, we use page_private(page).
403 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
406 if (!pfn_valid_within(page_to_pfn(buddy
)))
409 if (page_zone_id(page
) != page_zone_id(buddy
))
412 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
413 BUG_ON(page_count(buddy
) != 0);
420 * Freeing function for a buddy system allocator.
422 * The concept of a buddy system is to maintain direct-mapped table
423 * (containing bit values) for memory blocks of various "orders".
424 * The bottom level table contains the map for the smallest allocatable
425 * units of memory (here, pages), and each level above it describes
426 * pairs of units from the levels below, hence, "buddies".
427 * At a high level, all that happens here is marking the table entry
428 * at the bottom level available, and propagating the changes upward
429 * as necessary, plus some accounting needed to play nicely with other
430 * parts of the VM system.
431 * At each level, we keep a list of pages, which are heads of continuous
432 * free pages of length of (1 << order) and marked with PG_buddy. Page's
433 * order is recorded in page_private(page) field.
434 * So when we are allocating or freeing one, we can derive the state of the
435 * other. That is, if we allocate a small block, and both were
436 * free, the remainder of the region must be split into blocks.
437 * If a block is freed, and its buddy is also free, then this
438 * triggers coalescing into a block of larger size.
443 static inline void __free_one_page(struct page
*page
,
444 struct zone
*zone
, unsigned int order
)
446 unsigned long page_idx
;
447 int order_size
= 1 << order
;
448 int migratetype
= get_pageblock_migratetype(page
);
450 if (unlikely(PageCompound(page
)))
451 destroy_compound_page(page
, order
);
453 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
455 VM_BUG_ON(page_idx
& (order_size
- 1));
456 VM_BUG_ON(bad_range(zone
, page
));
458 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
459 while (order
< MAX_ORDER
-1) {
460 unsigned long combined_idx
;
463 buddy
= __page_find_buddy(page
, page_idx
, order
);
464 if (!page_is_buddy(page
, buddy
, order
))
465 break; /* Move the buddy up one level. */
467 list_del(&buddy
->lru
);
468 zone
->free_area
[order
].nr_free
--;
469 rmv_page_order(buddy
);
470 combined_idx
= __find_combined_index(page_idx
, order
);
471 page
= page
+ (combined_idx
- page_idx
);
472 page_idx
= combined_idx
;
475 set_page_order(page
, order
);
477 &zone
->free_area
[order
].free_list
[migratetype
]);
478 zone
->free_area
[order
].nr_free
++;
481 static inline int free_pages_check(struct page
*page
)
483 if (unlikely(page_mapcount(page
) |
484 (page
->mapping
!= NULL
) |
485 (page_count(page
) != 0) |
498 __ClearPageDirty(page
);
500 * For now, we report if PG_reserved was found set, but do not
501 * clear it, and do not free the page. But we shall soon need
502 * to do more, for when the ZERO_PAGE count wraps negative.
504 return PageReserved(page
);
508 * Frees a list of pages.
509 * Assumes all pages on list are in same zone, and of same order.
510 * count is the number of pages to free.
512 * If the zone was previously in an "all pages pinned" state then look to
513 * see if this freeing clears that state.
515 * And clear the zone's pages_scanned counter, to hold off the "all pages are
516 * pinned" detection logic.
518 static void free_pages_bulk(struct zone
*zone
, int count
,
519 struct list_head
*list
, int order
)
521 spin_lock(&zone
->lock
);
522 zone
->all_unreclaimable
= 0;
523 zone
->pages_scanned
= 0;
527 VM_BUG_ON(list_empty(list
));
528 page
= list_entry(list
->prev
, struct page
, lru
);
529 /* have to delete it as __free_one_page list manipulates */
530 list_del(&page
->lru
);
531 __free_one_page(page
, zone
, order
);
533 spin_unlock(&zone
->lock
);
536 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
538 spin_lock(&zone
->lock
);
539 zone
->all_unreclaimable
= 0;
540 zone
->pages_scanned
= 0;
541 __free_one_page(page
, zone
, order
);
542 spin_unlock(&zone
->lock
);
545 static void __free_pages_ok(struct page
*page
, unsigned int order
)
551 for (i
= 0 ; i
< (1 << order
) ; ++i
)
552 reserved
+= free_pages_check(page
+ i
);
556 if (!PageHighMem(page
))
557 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
558 arch_free_page(page
, order
);
559 kernel_map_pages(page
, 1 << order
, 0);
561 local_irq_save(flags
);
562 __count_vm_events(PGFREE
, 1 << order
);
563 free_one_page(page_zone(page
), page
, order
);
564 local_irq_restore(flags
);
568 * permit the bootmem allocator to evade page validation on high-order frees
570 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
573 __ClearPageReserved(page
);
574 set_page_count(page
, 0);
575 set_page_refcounted(page
);
581 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
582 struct page
*p
= &page
[loop
];
584 if (loop
+ 1 < BITS_PER_LONG
)
586 __ClearPageReserved(p
);
587 set_page_count(p
, 0);
590 set_page_refcounted(page
);
591 __free_pages(page
, order
);
597 * The order of subdivision here is critical for the IO subsystem.
598 * Please do not alter this order without good reasons and regression
599 * testing. Specifically, as large blocks of memory are subdivided,
600 * the order in which smaller blocks are delivered depends on the order
601 * they're subdivided in this function. This is the primary factor
602 * influencing the order in which pages are delivered to the IO
603 * subsystem according to empirical testing, and this is also justified
604 * by considering the behavior of a buddy system containing a single
605 * large block of memory acted on by a series of small allocations.
606 * This behavior is a critical factor in sglist merging's success.
610 static inline void expand(struct zone
*zone
, struct page
*page
,
611 int low
, int high
, struct free_area
*area
,
614 unsigned long size
= 1 << high
;
620 VM_BUG_ON(bad_range(zone
, &page
[size
]));
621 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
623 set_page_order(&page
[size
], high
);
628 * This page is about to be returned from the page allocator
630 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
632 if (unlikely(page_mapcount(page
) |
633 (page
->mapping
!= NULL
) |
634 (page_count(page
) != 0) |
649 * For now, we report if PG_reserved was found set, but do not
650 * clear it, and do not allocate the page: as a safety net.
652 if (PageReserved(page
))
655 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
656 1 << PG_referenced
| 1 << PG_arch_1
|
657 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
658 set_page_private(page
, 0);
659 set_page_refcounted(page
);
661 arch_alloc_page(page
, order
);
662 kernel_map_pages(page
, 1 << order
, 1);
664 if (gfp_flags
& __GFP_ZERO
)
665 prep_zero_page(page
, order
, gfp_flags
);
667 if (order
&& (gfp_flags
& __GFP_COMP
))
668 prep_compound_page(page
, order
);
673 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
675 * This array describes the order lists are fallen back to when
676 * the free lists for the desirable migrate type are depleted
678 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
679 [MIGRATE_UNMOVABLE
] = { MIGRATE_MOVABLE
},
680 [MIGRATE_MOVABLE
] = { MIGRATE_UNMOVABLE
},
684 * Move the free pages in a range to the free lists of the requested type.
685 * Note that start_page and end_pages are not aligned in a MAX_ORDER_NR_PAGES
686 * boundary. If alignment is required, use move_freepages_block()
688 int move_freepages(struct zone
*zone
,
689 struct page
*start_page
, struct page
*end_page
,
694 int blocks_moved
= 0;
696 #ifndef CONFIG_HOLES_IN_ZONE
698 * page_zone is not safe to call in this context when
699 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
700 * anyway as we check zone boundaries in move_freepages_block().
701 * Remove at a later date when no bug reports exist related to
702 * CONFIG_PAGE_GROUP_BY_MOBILITY
704 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
707 for (page
= start_page
; page
<= end_page
;) {
708 if (!pfn_valid_within(page_to_pfn(page
))) {
713 if (!PageBuddy(page
)) {
718 order
= page_order(page
);
719 list_del(&page
->lru
);
721 &zone
->free_area
[order
].free_list
[migratetype
]);
729 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
731 unsigned long start_pfn
, end_pfn
;
732 struct page
*start_page
, *end_page
;
734 start_pfn
= page_to_pfn(page
);
735 start_pfn
= start_pfn
& ~(MAX_ORDER_NR_PAGES
-1);
736 start_page
= pfn_to_page(start_pfn
);
737 end_page
= start_page
+ MAX_ORDER_NR_PAGES
- 1;
738 end_pfn
= start_pfn
+ MAX_ORDER_NR_PAGES
- 1;
740 /* Do not cross zone boundaries */
741 if (start_pfn
< zone
->zone_start_pfn
)
743 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
746 return move_freepages(zone
, start_page
, end_page
, migratetype
);
749 /* Remove an element from the buddy allocator from the fallback list */
750 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
751 int start_migratetype
)
753 struct free_area
* area
;
758 /* Find the largest possible block of pages in the other list */
759 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
761 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
762 migratetype
= fallbacks
[start_migratetype
][i
];
764 area
= &(zone
->free_area
[current_order
]);
765 if (list_empty(&area
->free_list
[migratetype
]))
768 page
= list_entry(area
->free_list
[migratetype
].next
,
773 * If breaking a large block of pages, move all free
774 * pages to the preferred allocation list
776 if (unlikely(current_order
>= MAX_ORDER
/ 2)) {
777 migratetype
= start_migratetype
;
778 move_freepages_block(zone
, page
, migratetype
);
781 /* Remove the page from the freelists */
782 list_del(&page
->lru
);
783 rmv_page_order(page
);
784 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
787 if (current_order
== MAX_ORDER
- 1)
788 set_pageblock_migratetype(page
,
791 expand(zone
, page
, order
, current_order
, area
, migratetype
);
799 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
800 int start_migratetype
)
804 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
807 * Do the hard work of removing an element from the buddy allocator.
808 * Call me with the zone->lock already held.
810 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
813 struct free_area
* area
;
814 unsigned int current_order
;
817 /* Find a page of the appropriate size in the preferred list */
818 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
819 area
= &(zone
->free_area
[current_order
]);
820 if (list_empty(&area
->free_list
[migratetype
]))
823 page
= list_entry(area
->free_list
[migratetype
].next
,
825 list_del(&page
->lru
);
826 rmv_page_order(page
);
828 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
829 expand(zone
, page
, order
, current_order
, area
, migratetype
);
833 page
= __rmqueue_fallback(zone
, order
, migratetype
);
841 * Obtain a specified number of elements from the buddy allocator, all under
842 * a single hold of the lock, for efficiency. Add them to the supplied list.
843 * Returns the number of new pages which were placed at *list.
845 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
846 unsigned long count
, struct list_head
*list
,
851 spin_lock(&zone
->lock
);
852 for (i
= 0; i
< count
; ++i
) {
853 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
854 if (unlikely(page
== NULL
))
856 list_add(&page
->lru
, list
);
857 set_page_private(page
, migratetype
);
859 spin_unlock(&zone
->lock
);
865 * Called from the vmstat counter updater to drain pagesets of this
866 * currently executing processor on remote nodes after they have
869 * Note that this function must be called with the thread pinned to
870 * a single processor.
872 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
877 local_irq_save(flags
);
878 if (pcp
->count
>= pcp
->batch
)
879 to_drain
= pcp
->batch
;
881 to_drain
= pcp
->count
;
882 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
883 pcp
->count
-= to_drain
;
884 local_irq_restore(flags
);
888 static void __drain_pages(unsigned int cpu
)
894 for_each_zone(zone
) {
895 struct per_cpu_pageset
*pset
;
897 if (!populated_zone(zone
))
900 pset
= zone_pcp(zone
, cpu
);
901 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
902 struct per_cpu_pages
*pcp
;
905 local_irq_save(flags
);
906 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
908 local_irq_restore(flags
);
913 #ifdef CONFIG_HIBERNATION
915 void mark_free_pages(struct zone
*zone
)
917 unsigned long pfn
, max_zone_pfn
;
920 struct list_head
*curr
;
922 if (!zone
->spanned_pages
)
925 spin_lock_irqsave(&zone
->lock
, flags
);
927 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
928 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
929 if (pfn_valid(pfn
)) {
930 struct page
*page
= pfn_to_page(pfn
);
932 if (!swsusp_page_is_forbidden(page
))
933 swsusp_unset_page_free(page
);
936 for_each_migratetype_order(order
, t
) {
937 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
940 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
941 for (i
= 0; i
< (1UL << order
); i
++)
942 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
945 spin_unlock_irqrestore(&zone
->lock
, flags
);
947 #endif /* CONFIG_PM */
949 #if defined(CONFIG_HIBERNATION) || defined(CONFIG_PAGE_GROUP_BY_MOBILITY)
951 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
953 void drain_local_pages(void)
957 local_irq_save(flags
);
958 __drain_pages(smp_processor_id());
959 local_irq_restore(flags
);
962 void smp_drain_local_pages(void *arg
)
968 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
970 void drain_all_local_pages(void)
974 local_irq_save(flags
);
975 __drain_pages(smp_processor_id());
976 local_irq_restore(flags
);
978 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
981 void drain_all_local_pages(void) {}
982 #endif /* CONFIG_HIBERNATION || CONFIG_PAGE_GROUP_BY_MOBILITY */
985 * Free a 0-order page
987 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
989 struct zone
*zone
= page_zone(page
);
990 struct per_cpu_pages
*pcp
;
994 page
->mapping
= NULL
;
995 if (free_pages_check(page
))
998 if (!PageHighMem(page
))
999 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1000 arch_free_page(page
, 0);
1001 kernel_map_pages(page
, 1, 0);
1003 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1004 local_irq_save(flags
);
1005 __count_vm_event(PGFREE
);
1006 list_add(&page
->lru
, &pcp
->list
);
1007 set_page_private(page
, get_pageblock_migratetype(page
));
1009 if (pcp
->count
>= pcp
->high
) {
1010 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1011 pcp
->count
-= pcp
->batch
;
1013 local_irq_restore(flags
);
1017 void fastcall
free_hot_page(struct page
*page
)
1019 free_hot_cold_page(page
, 0);
1022 void fastcall
free_cold_page(struct page
*page
)
1024 free_hot_cold_page(page
, 1);
1028 * split_page takes a non-compound higher-order page, and splits it into
1029 * n (1<<order) sub-pages: page[0..n]
1030 * Each sub-page must be freed individually.
1032 * Note: this is probably too low level an operation for use in drivers.
1033 * Please consult with lkml before using this in your driver.
1035 void split_page(struct page
*page
, unsigned int order
)
1039 VM_BUG_ON(PageCompound(page
));
1040 VM_BUG_ON(!page_count(page
));
1041 for (i
= 1; i
< (1 << order
); i
++)
1042 set_page_refcounted(page
+ i
);
1046 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1047 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1050 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1051 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1053 unsigned long flags
;
1055 int cold
= !!(gfp_flags
& __GFP_COLD
);
1057 int migratetype
= gfpflags_to_migratetype(gfp_flags
);
1061 if (likely(order
== 0)) {
1062 struct per_cpu_pages
*pcp
;
1064 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1065 local_irq_save(flags
);
1067 pcp
->count
= rmqueue_bulk(zone
, 0,
1068 pcp
->batch
, &pcp
->list
, migratetype
);
1069 if (unlikely(!pcp
->count
))
1073 #ifdef CONFIG_PAGE_GROUP_BY_MOBILITY
1074 /* Find a page of the appropriate migrate type */
1075 list_for_each_entry(page
, &pcp
->list
, lru
)
1076 if (page_private(page
) == migratetype
)
1079 /* Allocate more to the pcp list if necessary */
1080 if (unlikely(&page
->lru
== &pcp
->list
)) {
1081 pcp
->count
+= rmqueue_bulk(zone
, 0,
1082 pcp
->batch
, &pcp
->list
, migratetype
);
1083 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1086 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1087 #endif /* CONFIG_PAGE_GROUP_BY_MOBILITY */
1089 list_del(&page
->lru
);
1092 spin_lock_irqsave(&zone
->lock
, flags
);
1093 page
= __rmqueue(zone
, order
, migratetype
);
1094 spin_unlock(&zone
->lock
);
1099 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1100 zone_statistics(zonelist
, zone
);
1101 local_irq_restore(flags
);
1104 VM_BUG_ON(bad_range(zone
, page
));
1105 if (prep_new_page(page
, order
, gfp_flags
))
1110 local_irq_restore(flags
);
1115 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1116 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1117 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1118 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1119 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1120 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1121 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1123 #ifdef CONFIG_FAIL_PAGE_ALLOC
1125 static struct fail_page_alloc_attr
{
1126 struct fault_attr attr
;
1128 u32 ignore_gfp_highmem
;
1129 u32 ignore_gfp_wait
;
1132 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1134 struct dentry
*ignore_gfp_highmem_file
;
1135 struct dentry
*ignore_gfp_wait_file
;
1136 struct dentry
*min_order_file
;
1138 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1140 } fail_page_alloc
= {
1141 .attr
= FAULT_ATTR_INITIALIZER
,
1142 .ignore_gfp_wait
= 1,
1143 .ignore_gfp_highmem
= 1,
1147 static int __init
setup_fail_page_alloc(char *str
)
1149 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1151 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1153 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1155 if (order
< fail_page_alloc
.min_order
)
1157 if (gfp_mask
& __GFP_NOFAIL
)
1159 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1161 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1164 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1167 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1169 static int __init
fail_page_alloc_debugfs(void)
1171 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1175 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1179 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1181 fail_page_alloc
.ignore_gfp_wait_file
=
1182 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1183 &fail_page_alloc
.ignore_gfp_wait
);
1185 fail_page_alloc
.ignore_gfp_highmem_file
=
1186 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1187 &fail_page_alloc
.ignore_gfp_highmem
);
1188 fail_page_alloc
.min_order_file
=
1189 debugfs_create_u32("min-order", mode
, dir
,
1190 &fail_page_alloc
.min_order
);
1192 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1193 !fail_page_alloc
.ignore_gfp_highmem_file
||
1194 !fail_page_alloc
.min_order_file
) {
1196 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1197 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1198 debugfs_remove(fail_page_alloc
.min_order_file
);
1199 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1205 late_initcall(fail_page_alloc_debugfs
);
1207 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1209 #else /* CONFIG_FAIL_PAGE_ALLOC */
1211 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1216 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1219 * Return 1 if free pages are above 'mark'. This takes into account the order
1220 * of the allocation.
1222 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1223 int classzone_idx
, int alloc_flags
)
1225 /* free_pages my go negative - that's OK */
1227 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1230 if (alloc_flags
& ALLOC_HIGH
)
1232 if (alloc_flags
& ALLOC_HARDER
)
1235 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1237 for (o
= 0; o
< order
; o
++) {
1238 /* At the next order, this order's pages become unavailable */
1239 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1241 /* Require fewer higher order pages to be free */
1244 if (free_pages
<= min
)
1252 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1253 * skip over zones that are not allowed by the cpuset, or that have
1254 * been recently (in last second) found to be nearly full. See further
1255 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1256 * that have to skip over alot of full or unallowed zones.
1258 * If the zonelist cache is present in the passed in zonelist, then
1259 * returns a pointer to the allowed node mask (either the current
1260 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1262 * If the zonelist cache is not available for this zonelist, does
1263 * nothing and returns NULL.
1265 * If the fullzones BITMAP in the zonelist cache is stale (more than
1266 * a second since last zap'd) then we zap it out (clear its bits.)
1268 * We hold off even calling zlc_setup, until after we've checked the
1269 * first zone in the zonelist, on the theory that most allocations will
1270 * be satisfied from that first zone, so best to examine that zone as
1271 * quickly as we can.
1273 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1275 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1276 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1278 zlc
= zonelist
->zlcache_ptr
;
1282 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1283 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1284 zlc
->last_full_zap
= jiffies
;
1287 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1288 &cpuset_current_mems_allowed
:
1289 &node_states
[N_HIGH_MEMORY
];
1290 return allowednodes
;
1294 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1295 * if it is worth looking at further for free memory:
1296 * 1) Check that the zone isn't thought to be full (doesn't have its
1297 * bit set in the zonelist_cache fullzones BITMAP).
1298 * 2) Check that the zones node (obtained from the zonelist_cache
1299 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1300 * Return true (non-zero) if zone is worth looking at further, or
1301 * else return false (zero) if it is not.
1303 * This check -ignores- the distinction between various watermarks,
1304 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1305 * found to be full for any variation of these watermarks, it will
1306 * be considered full for up to one second by all requests, unless
1307 * we are so low on memory on all allowed nodes that we are forced
1308 * into the second scan of the zonelist.
1310 * In the second scan we ignore this zonelist cache and exactly
1311 * apply the watermarks to all zones, even it is slower to do so.
1312 * We are low on memory in the second scan, and should leave no stone
1313 * unturned looking for a free page.
1315 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1316 nodemask_t
*allowednodes
)
1318 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1319 int i
; /* index of *z in zonelist zones */
1320 int n
; /* node that zone *z is on */
1322 zlc
= zonelist
->zlcache_ptr
;
1326 i
= z
- zonelist
->zones
;
1329 /* This zone is worth trying if it is allowed but not full */
1330 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1334 * Given 'z' scanning a zonelist, set the corresponding bit in
1335 * zlc->fullzones, so that subsequent attempts to allocate a page
1336 * from that zone don't waste time re-examining it.
1338 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1340 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1341 int i
; /* index of *z in zonelist zones */
1343 zlc
= zonelist
->zlcache_ptr
;
1347 i
= z
- zonelist
->zones
;
1349 set_bit(i
, zlc
->fullzones
);
1352 #else /* CONFIG_NUMA */
1354 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1359 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1360 nodemask_t
*allowednodes
)
1365 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1368 #endif /* CONFIG_NUMA */
1371 * get_page_from_freelist goes through the zonelist trying to allocate
1374 static struct page
*
1375 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1376 struct zonelist
*zonelist
, int alloc_flags
)
1379 struct page
*page
= NULL
;
1380 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1382 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1383 int zlc_active
= 0; /* set if using zonelist_cache */
1384 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1385 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1389 * Scan zonelist, looking for a zone with enough free.
1390 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1392 z
= zonelist
->zones
;
1396 * In NUMA, this could be a policy zonelist which contains
1397 * zones that may not be allowed by the current gfp_mask.
1398 * Check the zone is allowed by the current flags
1400 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1401 if (highest_zoneidx
== -1)
1402 highest_zoneidx
= gfp_zone(gfp_mask
);
1403 if (zone_idx(*z
) > highest_zoneidx
)
1407 if (NUMA_BUILD
&& zlc_active
&&
1408 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1411 if ((alloc_flags
& ALLOC_CPUSET
) &&
1412 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1415 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1417 if (alloc_flags
& ALLOC_WMARK_MIN
)
1418 mark
= zone
->pages_min
;
1419 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1420 mark
= zone
->pages_low
;
1422 mark
= zone
->pages_high
;
1423 if (!zone_watermark_ok(zone
, order
, mark
,
1424 classzone_idx
, alloc_flags
)) {
1425 if (!zone_reclaim_mode
||
1426 !zone_reclaim(zone
, gfp_mask
, order
))
1427 goto this_zone_full
;
1431 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1436 zlc_mark_zone_full(zonelist
, z
);
1438 if (NUMA_BUILD
&& !did_zlc_setup
) {
1439 /* we do zlc_setup after the first zone is tried */
1440 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1444 } while (*(++z
) != NULL
);
1446 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1447 /* Disable zlc cache for second zonelist scan */
1455 * This is the 'heart' of the zoned buddy allocator.
1457 struct page
* fastcall
1458 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1459 struct zonelist
*zonelist
)
1461 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1464 struct reclaim_state reclaim_state
;
1465 struct task_struct
*p
= current
;
1468 int did_some_progress
;
1470 might_sleep_if(wait
);
1472 if (should_fail_alloc_page(gfp_mask
, order
))
1476 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1478 if (unlikely(*z
== NULL
)) {
1480 * Happens if we have an empty zonelist as a result of
1481 * GFP_THISNODE being used on a memoryless node
1486 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1487 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1492 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1493 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1494 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1495 * using a larger set of nodes after it has established that the
1496 * allowed per node queues are empty and that nodes are
1499 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1502 for (z
= zonelist
->zones
; *z
; z
++)
1503 wakeup_kswapd(*z
, order
);
1506 * OK, we're below the kswapd watermark and have kicked background
1507 * reclaim. Now things get more complex, so set up alloc_flags according
1508 * to how we want to proceed.
1510 * The caller may dip into page reserves a bit more if the caller
1511 * cannot run direct reclaim, or if the caller has realtime scheduling
1512 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1513 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1515 alloc_flags
= ALLOC_WMARK_MIN
;
1516 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1517 alloc_flags
|= ALLOC_HARDER
;
1518 if (gfp_mask
& __GFP_HIGH
)
1519 alloc_flags
|= ALLOC_HIGH
;
1521 alloc_flags
|= ALLOC_CPUSET
;
1524 * Go through the zonelist again. Let __GFP_HIGH and allocations
1525 * coming from realtime tasks go deeper into reserves.
1527 * This is the last chance, in general, before the goto nopage.
1528 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1529 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1531 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1535 /* This allocation should allow future memory freeing. */
1538 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1539 && !in_interrupt()) {
1540 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1542 /* go through the zonelist yet again, ignoring mins */
1543 page
= get_page_from_freelist(gfp_mask
, order
,
1544 zonelist
, ALLOC_NO_WATERMARKS
);
1547 if (gfp_mask
& __GFP_NOFAIL
) {
1548 congestion_wait(WRITE
, HZ
/50);
1555 /* Atomic allocations - we can't balance anything */
1561 /* We now go into synchronous reclaim */
1562 cpuset_memory_pressure_bump();
1563 p
->flags
|= PF_MEMALLOC
;
1564 reclaim_state
.reclaimed_slab
= 0;
1565 p
->reclaim_state
= &reclaim_state
;
1567 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1569 p
->reclaim_state
= NULL
;
1570 p
->flags
&= ~PF_MEMALLOC
;
1575 drain_all_local_pages();
1577 if (likely(did_some_progress
)) {
1578 page
= get_page_from_freelist(gfp_mask
, order
,
1579 zonelist
, alloc_flags
);
1582 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1584 * Go through the zonelist yet one more time, keep
1585 * very high watermark here, this is only to catch
1586 * a parallel oom killing, we must fail if we're still
1587 * under heavy pressure.
1589 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1590 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1594 /* The OOM killer will not help higher order allocs so fail */
1595 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1598 out_of_memory(zonelist
, gfp_mask
, order
);
1603 * Don't let big-order allocations loop unless the caller explicitly
1604 * requests that. Wait for some write requests to complete then retry.
1606 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1607 * <= 3, but that may not be true in other implementations.
1610 if (!(gfp_mask
& __GFP_NORETRY
)) {
1611 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1612 (gfp_mask
& __GFP_REPEAT
))
1614 if (gfp_mask
& __GFP_NOFAIL
)
1618 congestion_wait(WRITE
, HZ
/50);
1623 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1624 printk(KERN_WARNING
"%s: page allocation failure."
1625 " order:%d, mode:0x%x\n",
1626 p
->comm
, order
, gfp_mask
);
1634 EXPORT_SYMBOL(__alloc_pages
);
1637 * Common helper functions.
1639 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1642 page
= alloc_pages(gfp_mask
, order
);
1645 return (unsigned long) page_address(page
);
1648 EXPORT_SYMBOL(__get_free_pages
);
1650 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1655 * get_zeroed_page() returns a 32-bit address, which cannot represent
1658 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1660 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1662 return (unsigned long) page_address(page
);
1666 EXPORT_SYMBOL(get_zeroed_page
);
1668 void __pagevec_free(struct pagevec
*pvec
)
1670 int i
= pagevec_count(pvec
);
1673 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1676 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1678 if (put_page_testzero(page
)) {
1680 free_hot_page(page
);
1682 __free_pages_ok(page
, order
);
1686 EXPORT_SYMBOL(__free_pages
);
1688 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1691 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1692 __free_pages(virt_to_page((void *)addr
), order
);
1696 EXPORT_SYMBOL(free_pages
);
1698 static unsigned int nr_free_zone_pages(int offset
)
1700 /* Just pick one node, since fallback list is circular */
1701 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1702 unsigned int sum
= 0;
1704 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1705 struct zone
**zonep
= zonelist
->zones
;
1708 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1709 unsigned long size
= zone
->present_pages
;
1710 unsigned long high
= zone
->pages_high
;
1719 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1721 unsigned int nr_free_buffer_pages(void)
1723 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1725 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1728 * Amount of free RAM allocatable within all zones
1730 unsigned int nr_free_pagecache_pages(void)
1732 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1735 static inline void show_node(struct zone
*zone
)
1738 printk("Node %d ", zone_to_nid(zone
));
1741 void si_meminfo(struct sysinfo
*val
)
1743 val
->totalram
= totalram_pages
;
1745 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1746 val
->bufferram
= nr_blockdev_pages();
1747 val
->totalhigh
= totalhigh_pages
;
1748 val
->freehigh
= nr_free_highpages();
1749 val
->mem_unit
= PAGE_SIZE
;
1752 EXPORT_SYMBOL(si_meminfo
);
1755 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1757 pg_data_t
*pgdat
= NODE_DATA(nid
);
1759 val
->totalram
= pgdat
->node_present_pages
;
1760 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1761 #ifdef CONFIG_HIGHMEM
1762 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1763 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1769 val
->mem_unit
= PAGE_SIZE
;
1773 #define K(x) ((x) << (PAGE_SHIFT-10))
1776 * Show free area list (used inside shift_scroll-lock stuff)
1777 * We also calculate the percentage fragmentation. We do this by counting the
1778 * memory on each free list with the exception of the first item on the list.
1780 void show_free_areas(void)
1785 for_each_zone(zone
) {
1786 if (!populated_zone(zone
))
1790 printk("%s per-cpu:\n", zone
->name
);
1792 for_each_online_cpu(cpu
) {
1793 struct per_cpu_pageset
*pageset
;
1795 pageset
= zone_pcp(zone
, cpu
);
1797 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1798 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1799 cpu
, pageset
->pcp
[0].high
,
1800 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1801 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1802 pageset
->pcp
[1].count
);
1806 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1807 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1808 global_page_state(NR_ACTIVE
),
1809 global_page_state(NR_INACTIVE
),
1810 global_page_state(NR_FILE_DIRTY
),
1811 global_page_state(NR_WRITEBACK
),
1812 global_page_state(NR_UNSTABLE_NFS
),
1813 global_page_state(NR_FREE_PAGES
),
1814 global_page_state(NR_SLAB_RECLAIMABLE
) +
1815 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1816 global_page_state(NR_FILE_MAPPED
),
1817 global_page_state(NR_PAGETABLE
),
1818 global_page_state(NR_BOUNCE
));
1820 for_each_zone(zone
) {
1823 if (!populated_zone(zone
))
1835 " pages_scanned:%lu"
1836 " all_unreclaimable? %s"
1839 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1842 K(zone
->pages_high
),
1843 K(zone_page_state(zone
, NR_ACTIVE
)),
1844 K(zone_page_state(zone
, NR_INACTIVE
)),
1845 K(zone
->present_pages
),
1846 zone
->pages_scanned
,
1847 (zone
->all_unreclaimable
? "yes" : "no")
1849 printk("lowmem_reserve[]:");
1850 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1851 printk(" %lu", zone
->lowmem_reserve
[i
]);
1855 for_each_zone(zone
) {
1856 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1858 if (!populated_zone(zone
))
1862 printk("%s: ", zone
->name
);
1864 spin_lock_irqsave(&zone
->lock
, flags
);
1865 for (order
= 0; order
< MAX_ORDER
; order
++) {
1866 nr
[order
] = zone
->free_area
[order
].nr_free
;
1867 total
+= nr
[order
] << order
;
1869 spin_unlock_irqrestore(&zone
->lock
, flags
);
1870 for (order
= 0; order
< MAX_ORDER
; order
++)
1871 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1872 printk("= %lukB\n", K(total
));
1875 show_swap_cache_info();
1879 * Builds allocation fallback zone lists.
1881 * Add all populated zones of a node to the zonelist.
1883 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1884 int nr_zones
, enum zone_type zone_type
)
1888 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1893 zone
= pgdat
->node_zones
+ zone_type
;
1894 if (populated_zone(zone
)) {
1895 zonelist
->zones
[nr_zones
++] = zone
;
1896 check_highest_zone(zone_type
);
1899 } while (zone_type
);
1906 * 0 = automatic detection of better ordering.
1907 * 1 = order by ([node] distance, -zonetype)
1908 * 2 = order by (-zonetype, [node] distance)
1910 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1911 * the same zonelist. So only NUMA can configure this param.
1913 #define ZONELIST_ORDER_DEFAULT 0
1914 #define ZONELIST_ORDER_NODE 1
1915 #define ZONELIST_ORDER_ZONE 2
1917 /* zonelist order in the kernel.
1918 * set_zonelist_order() will set this to NODE or ZONE.
1920 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1921 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1925 /* The value user specified ....changed by config */
1926 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1927 /* string for sysctl */
1928 #define NUMA_ZONELIST_ORDER_LEN 16
1929 char numa_zonelist_order
[16] = "default";
1932 * interface for configure zonelist ordering.
1933 * command line option "numa_zonelist_order"
1934 * = "[dD]efault - default, automatic configuration.
1935 * = "[nN]ode - order by node locality, then by zone within node
1936 * = "[zZ]one - order by zone, then by locality within zone
1939 static int __parse_numa_zonelist_order(char *s
)
1941 if (*s
== 'd' || *s
== 'D') {
1942 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1943 } else if (*s
== 'n' || *s
== 'N') {
1944 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1945 } else if (*s
== 'z' || *s
== 'Z') {
1946 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1949 "Ignoring invalid numa_zonelist_order value: "
1956 static __init
int setup_numa_zonelist_order(char *s
)
1959 return __parse_numa_zonelist_order(s
);
1962 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1965 * sysctl handler for numa_zonelist_order
1967 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1968 struct file
*file
, void __user
*buffer
, size_t *length
,
1971 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1975 strncpy(saved_string
, (char*)table
->data
,
1976 NUMA_ZONELIST_ORDER_LEN
);
1977 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1981 int oldval
= user_zonelist_order
;
1982 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1984 * bogus value. restore saved string
1986 strncpy((char*)table
->data
, saved_string
,
1987 NUMA_ZONELIST_ORDER_LEN
);
1988 user_zonelist_order
= oldval
;
1989 } else if (oldval
!= user_zonelist_order
)
1990 build_all_zonelists();
1996 #define MAX_NODE_LOAD (num_online_nodes())
1997 static int node_load
[MAX_NUMNODES
];
2000 * find_next_best_node - find the next node that should appear in a given node's fallback list
2001 * @node: node whose fallback list we're appending
2002 * @used_node_mask: nodemask_t of already used nodes
2004 * We use a number of factors to determine which is the next node that should
2005 * appear on a given node's fallback list. The node should not have appeared
2006 * already in @node's fallback list, and it should be the next closest node
2007 * according to the distance array (which contains arbitrary distance values
2008 * from each node to each node in the system), and should also prefer nodes
2009 * with no CPUs, since presumably they'll have very little allocation pressure
2010 * on them otherwise.
2011 * It returns -1 if no node is found.
2013 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2016 int min_val
= INT_MAX
;
2019 /* Use the local node if we haven't already */
2020 if (!node_isset(node
, *used_node_mask
)) {
2021 node_set(node
, *used_node_mask
);
2025 for_each_node_state(n
, N_HIGH_MEMORY
) {
2028 /* Don't want a node to appear more than once */
2029 if (node_isset(n
, *used_node_mask
))
2032 /* Use the distance array to find the distance */
2033 val
= node_distance(node
, n
);
2035 /* Penalize nodes under us ("prefer the next node") */
2038 /* Give preference to headless and unused nodes */
2039 tmp
= node_to_cpumask(n
);
2040 if (!cpus_empty(tmp
))
2041 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2043 /* Slight preference for less loaded node */
2044 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2045 val
+= node_load
[n
];
2047 if (val
< min_val
) {
2054 node_set(best_node
, *used_node_mask
);
2061 * Build zonelists ordered by node and zones within node.
2062 * This results in maximum locality--normal zone overflows into local
2063 * DMA zone, if any--but risks exhausting DMA zone.
2065 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2069 struct zonelist
*zonelist
;
2071 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2072 zonelist
= pgdat
->node_zonelists
+ i
;
2073 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2075 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2076 zonelist
->zones
[j
] = NULL
;
2081 * Build gfp_thisnode zonelists
2083 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2087 struct zonelist
*zonelist
;
2089 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2090 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2091 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2092 zonelist
->zones
[j
] = NULL
;
2097 * Build zonelists ordered by zone and nodes within zones.
2098 * This results in conserving DMA zone[s] until all Normal memory is
2099 * exhausted, but results in overflowing to remote node while memory
2100 * may still exist in local DMA zone.
2102 static int node_order
[MAX_NUMNODES
];
2104 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2108 int zone_type
; /* needs to be signed */
2110 struct zonelist
*zonelist
;
2112 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2113 zonelist
= pgdat
->node_zonelists
+ i
;
2115 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2116 for (j
= 0; j
< nr_nodes
; j
++) {
2117 node
= node_order
[j
];
2118 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2119 if (populated_zone(z
)) {
2120 zonelist
->zones
[pos
++] = z
;
2121 check_highest_zone(zone_type
);
2125 zonelist
->zones
[pos
] = NULL
;
2129 static int default_zonelist_order(void)
2132 unsigned long low_kmem_size
,total_size
;
2136 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2137 * If they are really small and used heavily, the system can fall
2138 * into OOM very easily.
2139 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2141 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2144 for_each_online_node(nid
) {
2145 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2146 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2147 if (populated_zone(z
)) {
2148 if (zone_type
< ZONE_NORMAL
)
2149 low_kmem_size
+= z
->present_pages
;
2150 total_size
+= z
->present_pages
;
2154 if (!low_kmem_size
|| /* there are no DMA area. */
2155 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2156 return ZONELIST_ORDER_NODE
;
2158 * look into each node's config.
2159 * If there is a node whose DMA/DMA32 memory is very big area on
2160 * local memory, NODE_ORDER may be suitable.
2162 average_size
= total_size
/
2163 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2164 for_each_online_node(nid
) {
2167 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2168 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2169 if (populated_zone(z
)) {
2170 if (zone_type
< ZONE_NORMAL
)
2171 low_kmem_size
+= z
->present_pages
;
2172 total_size
+= z
->present_pages
;
2175 if (low_kmem_size
&&
2176 total_size
> average_size
&& /* ignore small node */
2177 low_kmem_size
> total_size
* 70/100)
2178 return ZONELIST_ORDER_NODE
;
2180 return ZONELIST_ORDER_ZONE
;
2183 static void set_zonelist_order(void)
2185 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2186 current_zonelist_order
= default_zonelist_order();
2188 current_zonelist_order
= user_zonelist_order
;
2191 static void build_zonelists(pg_data_t
*pgdat
)
2195 nodemask_t used_mask
;
2196 int local_node
, prev_node
;
2197 struct zonelist
*zonelist
;
2198 int order
= current_zonelist_order
;
2200 /* initialize zonelists */
2201 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2202 zonelist
= pgdat
->node_zonelists
+ i
;
2203 zonelist
->zones
[0] = NULL
;
2206 /* NUMA-aware ordering of nodes */
2207 local_node
= pgdat
->node_id
;
2208 load
= num_online_nodes();
2209 prev_node
= local_node
;
2210 nodes_clear(used_mask
);
2212 memset(node_load
, 0, sizeof(node_load
));
2213 memset(node_order
, 0, sizeof(node_order
));
2216 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2217 int distance
= node_distance(local_node
, node
);
2220 * If another node is sufficiently far away then it is better
2221 * to reclaim pages in a zone before going off node.
2223 if (distance
> RECLAIM_DISTANCE
)
2224 zone_reclaim_mode
= 1;
2227 * We don't want to pressure a particular node.
2228 * So adding penalty to the first node in same
2229 * distance group to make it round-robin.
2231 if (distance
!= node_distance(local_node
, prev_node
))
2232 node_load
[node
] = load
;
2236 if (order
== ZONELIST_ORDER_NODE
)
2237 build_zonelists_in_node_order(pgdat
, node
);
2239 node_order
[j
++] = node
; /* remember order */
2242 if (order
== ZONELIST_ORDER_ZONE
) {
2243 /* calculate node order -- i.e., DMA last! */
2244 build_zonelists_in_zone_order(pgdat
, j
);
2247 build_thisnode_zonelists(pgdat
);
2250 /* Construct the zonelist performance cache - see further mmzone.h */
2251 static void build_zonelist_cache(pg_data_t
*pgdat
)
2255 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2256 struct zonelist
*zonelist
;
2257 struct zonelist_cache
*zlc
;
2260 zonelist
= pgdat
->node_zonelists
+ i
;
2261 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2262 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2263 for (z
= zonelist
->zones
; *z
; z
++)
2264 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2269 #else /* CONFIG_NUMA */
2271 static void set_zonelist_order(void)
2273 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2276 static void build_zonelists(pg_data_t
*pgdat
)
2278 int node
, local_node
;
2281 local_node
= pgdat
->node_id
;
2282 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2283 struct zonelist
*zonelist
;
2285 zonelist
= pgdat
->node_zonelists
+ i
;
2287 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2289 * Now we build the zonelist so that it contains the zones
2290 * of all the other nodes.
2291 * We don't want to pressure a particular node, so when
2292 * building the zones for node N, we make sure that the
2293 * zones coming right after the local ones are those from
2294 * node N+1 (modulo N)
2296 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2297 if (!node_online(node
))
2299 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2301 for (node
= 0; node
< local_node
; node
++) {
2302 if (!node_online(node
))
2304 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2307 zonelist
->zones
[j
] = NULL
;
2311 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2312 static void build_zonelist_cache(pg_data_t
*pgdat
)
2316 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2317 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2320 #endif /* CONFIG_NUMA */
2322 /* return values int ....just for stop_machine_run() */
2323 static int __build_all_zonelists(void *dummy
)
2327 for_each_online_node(nid
) {
2328 pg_data_t
*pgdat
= NODE_DATA(nid
);
2330 build_zonelists(pgdat
);
2331 build_zonelist_cache(pgdat
);
2336 void build_all_zonelists(void)
2338 set_zonelist_order();
2340 if (system_state
== SYSTEM_BOOTING
) {
2341 __build_all_zonelists(NULL
);
2342 cpuset_init_current_mems_allowed();
2344 /* we have to stop all cpus to guaranntee there is no user
2346 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2347 /* cpuset refresh routine should be here */
2349 vm_total_pages
= nr_free_pagecache_pages();
2350 printk("Built %i zonelists in %s order. Total pages: %ld\n",
2352 zonelist_order_name
[current_zonelist_order
],
2355 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2360 * Helper functions to size the waitqueue hash table.
2361 * Essentially these want to choose hash table sizes sufficiently
2362 * large so that collisions trying to wait on pages are rare.
2363 * But in fact, the number of active page waitqueues on typical
2364 * systems is ridiculously low, less than 200. So this is even
2365 * conservative, even though it seems large.
2367 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2368 * waitqueues, i.e. the size of the waitq table given the number of pages.
2370 #define PAGES_PER_WAITQUEUE 256
2372 #ifndef CONFIG_MEMORY_HOTPLUG
2373 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2375 unsigned long size
= 1;
2377 pages
/= PAGES_PER_WAITQUEUE
;
2379 while (size
< pages
)
2383 * Once we have dozens or even hundreds of threads sleeping
2384 * on IO we've got bigger problems than wait queue collision.
2385 * Limit the size of the wait table to a reasonable size.
2387 size
= min(size
, 4096UL);
2389 return max(size
, 4UL);
2393 * A zone's size might be changed by hot-add, so it is not possible to determine
2394 * a suitable size for its wait_table. So we use the maximum size now.
2396 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2398 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2399 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2400 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2402 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2403 * or more by the traditional way. (See above). It equals:
2405 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2406 * ia64(16K page size) : = ( 8G + 4M)byte.
2407 * powerpc (64K page size) : = (32G +16M)byte.
2409 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2416 * This is an integer logarithm so that shifts can be used later
2417 * to extract the more random high bits from the multiplicative
2418 * hash function before the remainder is taken.
2420 static inline unsigned long wait_table_bits(unsigned long size
)
2425 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2428 * Initially all pages are reserved - free ones are freed
2429 * up by free_all_bootmem() once the early boot process is
2430 * done. Non-atomic initialization, single-pass.
2432 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2433 unsigned long start_pfn
, enum memmap_context context
)
2436 unsigned long end_pfn
= start_pfn
+ size
;
2439 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2441 * There can be holes in boot-time mem_map[]s
2442 * handed to this function. They do not
2443 * exist on hotplugged memory.
2445 if (context
== MEMMAP_EARLY
) {
2446 if (!early_pfn_valid(pfn
))
2448 if (!early_pfn_in_nid(pfn
, nid
))
2451 page
= pfn_to_page(pfn
);
2452 set_page_links(page
, zone
, nid
, pfn
);
2453 init_page_count(page
);
2454 reset_page_mapcount(page
);
2455 SetPageReserved(page
);
2458 * Mark the block movable so that blocks are reserved for
2459 * movable at startup. This will force kernel allocations
2460 * to reserve their blocks rather than leaking throughout
2461 * the address space during boot when many long-lived
2462 * kernel allocations are made
2464 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2466 INIT_LIST_HEAD(&page
->lru
);
2467 #ifdef WANT_PAGE_VIRTUAL
2468 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2469 if (!is_highmem_idx(zone
))
2470 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2475 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2476 struct zone
*zone
, unsigned long size
)
2479 for_each_migratetype_order(order
, t
) {
2480 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2481 zone
->free_area
[order
].nr_free
= 0;
2485 #ifndef __HAVE_ARCH_MEMMAP_INIT
2486 #define memmap_init(size, nid, zone, start_pfn) \
2487 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2490 static int __devinit
zone_batchsize(struct zone
*zone
)
2495 * The per-cpu-pages pools are set to around 1000th of the
2496 * size of the zone. But no more than 1/2 of a meg.
2498 * OK, so we don't know how big the cache is. So guess.
2500 batch
= zone
->present_pages
/ 1024;
2501 if (batch
* PAGE_SIZE
> 512 * 1024)
2502 batch
= (512 * 1024) / PAGE_SIZE
;
2503 batch
/= 4; /* We effectively *= 4 below */
2508 * Clamp the batch to a 2^n - 1 value. Having a power
2509 * of 2 value was found to be more likely to have
2510 * suboptimal cache aliasing properties in some cases.
2512 * For example if 2 tasks are alternately allocating
2513 * batches of pages, one task can end up with a lot
2514 * of pages of one half of the possible page colors
2515 * and the other with pages of the other colors.
2517 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2522 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2524 struct per_cpu_pages
*pcp
;
2526 memset(p
, 0, sizeof(*p
));
2528 pcp
= &p
->pcp
[0]; /* hot */
2530 pcp
->high
= 6 * batch
;
2531 pcp
->batch
= max(1UL, 1 * batch
);
2532 INIT_LIST_HEAD(&pcp
->list
);
2534 pcp
= &p
->pcp
[1]; /* cold*/
2536 pcp
->high
= 2 * batch
;
2537 pcp
->batch
= max(1UL, batch
/2);
2538 INIT_LIST_HEAD(&pcp
->list
);
2542 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2543 * to the value high for the pageset p.
2546 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2549 struct per_cpu_pages
*pcp
;
2551 pcp
= &p
->pcp
[0]; /* hot list */
2553 pcp
->batch
= max(1UL, high
/4);
2554 if ((high
/4) > (PAGE_SHIFT
* 8))
2555 pcp
->batch
= PAGE_SHIFT
* 8;
2561 * Boot pageset table. One per cpu which is going to be used for all
2562 * zones and all nodes. The parameters will be set in such a way
2563 * that an item put on a list will immediately be handed over to
2564 * the buddy list. This is safe since pageset manipulation is done
2565 * with interrupts disabled.
2567 * Some NUMA counter updates may also be caught by the boot pagesets.
2569 * The boot_pagesets must be kept even after bootup is complete for
2570 * unused processors and/or zones. They do play a role for bootstrapping
2571 * hotplugged processors.
2573 * zoneinfo_show() and maybe other functions do
2574 * not check if the processor is online before following the pageset pointer.
2575 * Other parts of the kernel may not check if the zone is available.
2577 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2580 * Dynamically allocate memory for the
2581 * per cpu pageset array in struct zone.
2583 static int __cpuinit
process_zones(int cpu
)
2585 struct zone
*zone
, *dzone
;
2586 int node
= cpu_to_node(cpu
);
2588 node_set_state(node
, N_CPU
); /* this node has a cpu */
2590 for_each_zone(zone
) {
2592 if (!populated_zone(zone
))
2595 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2597 if (!zone_pcp(zone
, cpu
))
2600 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2602 if (percpu_pagelist_fraction
)
2603 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2604 (zone
->present_pages
/ percpu_pagelist_fraction
));
2609 for_each_zone(dzone
) {
2610 if (!populated_zone(dzone
))
2614 kfree(zone_pcp(dzone
, cpu
));
2615 zone_pcp(dzone
, cpu
) = NULL
;
2620 static inline void free_zone_pagesets(int cpu
)
2624 for_each_zone(zone
) {
2625 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2627 /* Free per_cpu_pageset if it is slab allocated */
2628 if (pset
!= &boot_pageset
[cpu
])
2630 zone_pcp(zone
, cpu
) = NULL
;
2634 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2635 unsigned long action
,
2638 int cpu
= (long)hcpu
;
2639 int ret
= NOTIFY_OK
;
2642 case CPU_UP_PREPARE
:
2643 case CPU_UP_PREPARE_FROZEN
:
2644 if (process_zones(cpu
))
2647 case CPU_UP_CANCELED
:
2648 case CPU_UP_CANCELED_FROZEN
:
2650 case CPU_DEAD_FROZEN
:
2651 free_zone_pagesets(cpu
);
2659 static struct notifier_block __cpuinitdata pageset_notifier
=
2660 { &pageset_cpuup_callback
, NULL
, 0 };
2662 void __init
setup_per_cpu_pageset(void)
2666 /* Initialize per_cpu_pageset for cpu 0.
2667 * A cpuup callback will do this for every cpu
2668 * as it comes online
2670 err
= process_zones(smp_processor_id());
2672 register_cpu_notifier(&pageset_notifier
);
2677 static noinline __init_refok
2678 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2681 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2685 * The per-page waitqueue mechanism uses hashed waitqueues
2688 zone
->wait_table_hash_nr_entries
=
2689 wait_table_hash_nr_entries(zone_size_pages
);
2690 zone
->wait_table_bits
=
2691 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2692 alloc_size
= zone
->wait_table_hash_nr_entries
2693 * sizeof(wait_queue_head_t
);
2695 if (system_state
== SYSTEM_BOOTING
) {
2696 zone
->wait_table
= (wait_queue_head_t
*)
2697 alloc_bootmem_node(pgdat
, alloc_size
);
2700 * This case means that a zone whose size was 0 gets new memory
2701 * via memory hot-add.
2702 * But it may be the case that a new node was hot-added. In
2703 * this case vmalloc() will not be able to use this new node's
2704 * memory - this wait_table must be initialized to use this new
2705 * node itself as well.
2706 * To use this new node's memory, further consideration will be
2709 zone
->wait_table
= vmalloc(alloc_size
);
2711 if (!zone
->wait_table
)
2714 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2715 init_waitqueue_head(zone
->wait_table
+ i
);
2720 static __meminit
void zone_pcp_init(struct zone
*zone
)
2723 unsigned long batch
= zone_batchsize(zone
);
2725 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2727 /* Early boot. Slab allocator not functional yet */
2728 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2729 setup_pageset(&boot_pageset
[cpu
],0);
2731 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2734 if (zone
->present_pages
)
2735 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2736 zone
->name
, zone
->present_pages
, batch
);
2739 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2740 unsigned long zone_start_pfn
,
2742 enum memmap_context context
)
2744 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2746 ret
= zone_wait_table_init(zone
, size
);
2749 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2751 zone
->zone_start_pfn
= zone_start_pfn
;
2753 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2755 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2760 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2762 * Basic iterator support. Return the first range of PFNs for a node
2763 * Note: nid == MAX_NUMNODES returns first region regardless of node
2765 static int __meminit
first_active_region_index_in_nid(int nid
)
2769 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2770 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2777 * Basic iterator support. Return the next active range of PFNs for a node
2778 * Note: nid == MAX_NUMNODES returns next region regardles of node
2780 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2782 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2783 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2789 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2791 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2792 * Architectures may implement their own version but if add_active_range()
2793 * was used and there are no special requirements, this is a convenient
2796 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2800 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2801 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2802 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2804 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2805 return early_node_map
[i
].nid
;
2810 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2812 /* Basic iterator support to walk early_node_map[] */
2813 #define for_each_active_range_index_in_nid(i, nid) \
2814 for (i = first_active_region_index_in_nid(nid); i != -1; \
2815 i = next_active_region_index_in_nid(i, nid))
2818 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2819 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2820 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2822 * If an architecture guarantees that all ranges registered with
2823 * add_active_ranges() contain no holes and may be freed, this
2824 * this function may be used instead of calling free_bootmem() manually.
2826 void __init
free_bootmem_with_active_regions(int nid
,
2827 unsigned long max_low_pfn
)
2831 for_each_active_range_index_in_nid(i
, nid
) {
2832 unsigned long size_pages
= 0;
2833 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2835 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2838 if (end_pfn
> max_low_pfn
)
2839 end_pfn
= max_low_pfn
;
2841 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2842 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2843 PFN_PHYS(early_node_map
[i
].start_pfn
),
2844 size_pages
<< PAGE_SHIFT
);
2849 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2850 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2852 * If an architecture guarantees that all ranges registered with
2853 * add_active_ranges() contain no holes and may be freed, this
2854 * function may be used instead of calling memory_present() manually.
2856 void __init
sparse_memory_present_with_active_regions(int nid
)
2860 for_each_active_range_index_in_nid(i
, nid
)
2861 memory_present(early_node_map
[i
].nid
,
2862 early_node_map
[i
].start_pfn
,
2863 early_node_map
[i
].end_pfn
);
2867 * push_node_boundaries - Push node boundaries to at least the requested boundary
2868 * @nid: The nid of the node to push the boundary for
2869 * @start_pfn: The start pfn of the node
2870 * @end_pfn: The end pfn of the node
2872 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2873 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2874 * be hotplugged even though no physical memory exists. This function allows
2875 * an arch to push out the node boundaries so mem_map is allocated that can
2878 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2879 void __init
push_node_boundaries(unsigned int nid
,
2880 unsigned long start_pfn
, unsigned long end_pfn
)
2882 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2883 nid
, start_pfn
, end_pfn
);
2885 /* Initialise the boundary for this node if necessary */
2886 if (node_boundary_end_pfn
[nid
] == 0)
2887 node_boundary_start_pfn
[nid
] = -1UL;
2889 /* Update the boundaries */
2890 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2891 node_boundary_start_pfn
[nid
] = start_pfn
;
2892 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2893 node_boundary_end_pfn
[nid
] = end_pfn
;
2896 /* If necessary, push the node boundary out for reserve hotadd */
2897 static void __meminit
account_node_boundary(unsigned int nid
,
2898 unsigned long *start_pfn
, unsigned long *end_pfn
)
2900 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2901 nid
, *start_pfn
, *end_pfn
);
2903 /* Return if boundary information has not been provided */
2904 if (node_boundary_end_pfn
[nid
] == 0)
2907 /* Check the boundaries and update if necessary */
2908 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2909 *start_pfn
= node_boundary_start_pfn
[nid
];
2910 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2911 *end_pfn
= node_boundary_end_pfn
[nid
];
2914 void __init
push_node_boundaries(unsigned int nid
,
2915 unsigned long start_pfn
, unsigned long end_pfn
) {}
2917 static void __meminit
account_node_boundary(unsigned int nid
,
2918 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
2923 * get_pfn_range_for_nid - Return the start and end page frames for a node
2924 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2925 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2926 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2928 * It returns the start and end page frame of a node based on information
2929 * provided by an arch calling add_active_range(). If called for a node
2930 * with no available memory, a warning is printed and the start and end
2933 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
2934 unsigned long *start_pfn
, unsigned long *end_pfn
)
2940 for_each_active_range_index_in_nid(i
, nid
) {
2941 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
2942 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
2945 if (*start_pfn
== -1UL)
2948 /* Push the node boundaries out if requested */
2949 account_node_boundary(nid
, start_pfn
, end_pfn
);
2953 * This finds a zone that can be used for ZONE_MOVABLE pages. The
2954 * assumption is made that zones within a node are ordered in monotonic
2955 * increasing memory addresses so that the "highest" populated zone is used
2957 void __init
find_usable_zone_for_movable(void)
2960 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
2961 if (zone_index
== ZONE_MOVABLE
)
2964 if (arch_zone_highest_possible_pfn
[zone_index
] >
2965 arch_zone_lowest_possible_pfn
[zone_index
])
2969 VM_BUG_ON(zone_index
== -1);
2970 movable_zone
= zone_index
;
2974 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
2975 * because it is sized independant of architecture. Unlike the other zones,
2976 * the starting point for ZONE_MOVABLE is not fixed. It may be different
2977 * in each node depending on the size of each node and how evenly kernelcore
2978 * is distributed. This helper function adjusts the zone ranges
2979 * provided by the architecture for a given node by using the end of the
2980 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
2981 * zones within a node are in order of monotonic increases memory addresses
2983 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
2984 unsigned long zone_type
,
2985 unsigned long node_start_pfn
,
2986 unsigned long node_end_pfn
,
2987 unsigned long *zone_start_pfn
,
2988 unsigned long *zone_end_pfn
)
2990 /* Only adjust if ZONE_MOVABLE is on this node */
2991 if (zone_movable_pfn
[nid
]) {
2992 /* Size ZONE_MOVABLE */
2993 if (zone_type
== ZONE_MOVABLE
) {
2994 *zone_start_pfn
= zone_movable_pfn
[nid
];
2995 *zone_end_pfn
= min(node_end_pfn
,
2996 arch_zone_highest_possible_pfn
[movable_zone
]);
2998 /* Adjust for ZONE_MOVABLE starting within this range */
2999 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3000 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3001 *zone_end_pfn
= zone_movable_pfn
[nid
];
3003 /* Check if this whole range is within ZONE_MOVABLE */
3004 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3005 *zone_start_pfn
= *zone_end_pfn
;
3010 * Return the number of pages a zone spans in a node, including holes
3011 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3013 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3014 unsigned long zone_type
,
3015 unsigned long *ignored
)
3017 unsigned long node_start_pfn
, node_end_pfn
;
3018 unsigned long zone_start_pfn
, zone_end_pfn
;
3020 /* Get the start and end of the node and zone */
3021 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3022 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3023 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3024 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3025 node_start_pfn
, node_end_pfn
,
3026 &zone_start_pfn
, &zone_end_pfn
);
3028 /* Check that this node has pages within the zone's required range */
3029 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3032 /* Move the zone boundaries inside the node if necessary */
3033 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3034 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3036 /* Return the spanned pages */
3037 return zone_end_pfn
- zone_start_pfn
;
3041 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3042 * then all holes in the requested range will be accounted for.
3044 unsigned long __meminit
__absent_pages_in_range(int nid
,
3045 unsigned long range_start_pfn
,
3046 unsigned long range_end_pfn
)
3049 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3050 unsigned long start_pfn
;
3052 /* Find the end_pfn of the first active range of pfns in the node */
3053 i
= first_active_region_index_in_nid(nid
);
3057 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3059 /* Account for ranges before physical memory on this node */
3060 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3061 hole_pages
= prev_end_pfn
- range_start_pfn
;
3063 /* Find all holes for the zone within the node */
3064 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3066 /* No need to continue if prev_end_pfn is outside the zone */
3067 if (prev_end_pfn
>= range_end_pfn
)
3070 /* Make sure the end of the zone is not within the hole */
3071 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3072 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3074 /* Update the hole size cound and move on */
3075 if (start_pfn
> range_start_pfn
) {
3076 BUG_ON(prev_end_pfn
> start_pfn
);
3077 hole_pages
+= start_pfn
- prev_end_pfn
;
3079 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3082 /* Account for ranges past physical memory on this node */
3083 if (range_end_pfn
> prev_end_pfn
)
3084 hole_pages
+= range_end_pfn
-
3085 max(range_start_pfn
, prev_end_pfn
);
3091 * absent_pages_in_range - Return number of page frames in holes within a range
3092 * @start_pfn: The start PFN to start searching for holes
3093 * @end_pfn: The end PFN to stop searching for holes
3095 * It returns the number of pages frames in memory holes within a range.
3097 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3098 unsigned long end_pfn
)
3100 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3103 /* Return the number of page frames in holes in a zone on a node */
3104 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3105 unsigned long zone_type
,
3106 unsigned long *ignored
)
3108 unsigned long node_start_pfn
, node_end_pfn
;
3109 unsigned long zone_start_pfn
, zone_end_pfn
;
3111 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3112 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3114 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3117 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3118 node_start_pfn
, node_end_pfn
,
3119 &zone_start_pfn
, &zone_end_pfn
);
3120 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3124 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3125 unsigned long zone_type
,
3126 unsigned long *zones_size
)
3128 return zones_size
[zone_type
];
3131 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3132 unsigned long zone_type
,
3133 unsigned long *zholes_size
)
3138 return zholes_size
[zone_type
];
3143 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3144 unsigned long *zones_size
, unsigned long *zholes_size
)
3146 unsigned long realtotalpages
, totalpages
= 0;
3149 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3150 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3152 pgdat
->node_spanned_pages
= totalpages
;
3154 realtotalpages
= totalpages
;
3155 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3157 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3159 pgdat
->node_present_pages
= realtotalpages
;
3160 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3164 #ifndef CONFIG_SPARSEMEM
3166 * Calculate the size of the zone->blockflags rounded to an unsigned long
3167 * Start by making sure zonesize is a multiple of MAX_ORDER-1 by rounding up
3168 * Then figure 1 NR_PAGEBLOCK_BITS worth of bits per MAX_ORDER-1, finally
3169 * round what is now in bits to nearest long in bits, then return it in
3172 static unsigned long __init
usemap_size(unsigned long zonesize
)
3174 unsigned long usemapsize
;
3176 usemapsize
= roundup(zonesize
, MAX_ORDER_NR_PAGES
);
3177 usemapsize
= usemapsize
>> (MAX_ORDER
-1);
3178 usemapsize
*= NR_PAGEBLOCK_BITS
;
3179 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3181 return usemapsize
/ 8;
3184 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3185 struct zone
*zone
, unsigned long zonesize
)
3187 unsigned long usemapsize
= usemap_size(zonesize
);
3188 zone
->pageblock_flags
= NULL
;
3190 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3191 memset(zone
->pageblock_flags
, 0, usemapsize
);
3195 static void inline setup_usemap(struct pglist_data
*pgdat
,
3196 struct zone
*zone
, unsigned long zonesize
) {}
3197 #endif /* CONFIG_SPARSEMEM */
3200 * Set up the zone data structures:
3201 * - mark all pages reserved
3202 * - mark all memory queues empty
3203 * - clear the memory bitmaps
3205 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3206 unsigned long *zones_size
, unsigned long *zholes_size
)
3209 int nid
= pgdat
->node_id
;
3210 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3213 pgdat_resize_init(pgdat
);
3214 pgdat
->nr_zones
= 0;
3215 init_waitqueue_head(&pgdat
->kswapd_wait
);
3216 pgdat
->kswapd_max_order
= 0;
3218 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3219 struct zone
*zone
= pgdat
->node_zones
+ j
;
3220 unsigned long size
, realsize
, memmap_pages
;
3222 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3223 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3227 * Adjust realsize so that it accounts for how much memory
3228 * is used by this zone for memmap. This affects the watermark
3229 * and per-cpu initialisations
3231 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3232 if (realsize
>= memmap_pages
) {
3233 realsize
-= memmap_pages
;
3235 " %s zone: %lu pages used for memmap\n",
3236 zone_names
[j
], memmap_pages
);
3239 " %s zone: %lu pages exceeds realsize %lu\n",
3240 zone_names
[j
], memmap_pages
, realsize
);
3242 /* Account for reserved pages */
3243 if (j
== 0 && realsize
> dma_reserve
) {
3244 realsize
-= dma_reserve
;
3245 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3246 zone_names
[0], dma_reserve
);
3249 if (!is_highmem_idx(j
))
3250 nr_kernel_pages
+= realsize
;
3251 nr_all_pages
+= realsize
;
3253 zone
->spanned_pages
= size
;
3254 zone
->present_pages
= realsize
;
3257 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3259 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3261 zone
->name
= zone_names
[j
];
3262 spin_lock_init(&zone
->lock
);
3263 spin_lock_init(&zone
->lru_lock
);
3264 zone_seqlock_init(zone
);
3265 zone
->zone_pgdat
= pgdat
;
3267 zone
->prev_priority
= DEF_PRIORITY
;
3269 zone_pcp_init(zone
);
3270 INIT_LIST_HEAD(&zone
->active_list
);
3271 INIT_LIST_HEAD(&zone
->inactive_list
);
3272 zone
->nr_scan_active
= 0;
3273 zone
->nr_scan_inactive
= 0;
3274 zap_zone_vm_stats(zone
);
3275 atomic_set(&zone
->reclaim_in_progress
, 0);
3279 setup_usemap(pgdat
, zone
, size
);
3280 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3281 size
, MEMMAP_EARLY
);
3283 zone_start_pfn
+= size
;
3287 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3289 /* Skip empty nodes */
3290 if (!pgdat
->node_spanned_pages
)
3293 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3294 /* ia64 gets its own node_mem_map, before this, without bootmem */
3295 if (!pgdat
->node_mem_map
) {
3296 unsigned long size
, start
, end
;
3300 * The zone's endpoints aren't required to be MAX_ORDER
3301 * aligned but the node_mem_map endpoints must be in order
3302 * for the buddy allocator to function correctly.
3304 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3305 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3306 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3307 size
= (end
- start
) * sizeof(struct page
);
3308 map
= alloc_remap(pgdat
->node_id
, size
);
3310 map
= alloc_bootmem_node(pgdat
, size
);
3311 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3313 #ifndef CONFIG_NEED_MULTIPLE_NODES
3315 * With no DISCONTIG, the global mem_map is just set as node 0's
3317 if (pgdat
== NODE_DATA(0)) {
3318 mem_map
= NODE_DATA(0)->node_mem_map
;
3319 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3320 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3321 mem_map
-= pgdat
->node_start_pfn
;
3322 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3325 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3328 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3329 unsigned long *zones_size
, unsigned long node_start_pfn
,
3330 unsigned long *zholes_size
)
3332 pgdat
->node_id
= nid
;
3333 pgdat
->node_start_pfn
= node_start_pfn
;
3334 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3336 alloc_node_mem_map(pgdat
);
3338 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3341 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3343 #if MAX_NUMNODES > 1
3345 * Figure out the number of possible node ids.
3347 static void __init
setup_nr_node_ids(void)
3350 unsigned int highest
= 0;
3352 for_each_node_mask(node
, node_possible_map
)
3354 nr_node_ids
= highest
+ 1;
3357 static inline void setup_nr_node_ids(void)
3363 * add_active_range - Register a range of PFNs backed by physical memory
3364 * @nid: The node ID the range resides on
3365 * @start_pfn: The start PFN of the available physical memory
3366 * @end_pfn: The end PFN of the available physical memory
3368 * These ranges are stored in an early_node_map[] and later used by
3369 * free_area_init_nodes() to calculate zone sizes and holes. If the
3370 * range spans a memory hole, it is up to the architecture to ensure
3371 * the memory is not freed by the bootmem allocator. If possible
3372 * the range being registered will be merged with existing ranges.
3374 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3375 unsigned long end_pfn
)
3379 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3380 "%d entries of %d used\n",
3381 nid
, start_pfn
, end_pfn
,
3382 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3384 /* Merge with existing active regions if possible */
3385 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3386 if (early_node_map
[i
].nid
!= nid
)
3389 /* Skip if an existing region covers this new one */
3390 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3391 end_pfn
<= early_node_map
[i
].end_pfn
)
3394 /* Merge forward if suitable */
3395 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3396 end_pfn
> early_node_map
[i
].end_pfn
) {
3397 early_node_map
[i
].end_pfn
= end_pfn
;
3401 /* Merge backward if suitable */
3402 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3403 end_pfn
>= early_node_map
[i
].start_pfn
) {
3404 early_node_map
[i
].start_pfn
= start_pfn
;
3409 /* Check that early_node_map is large enough */
3410 if (i
>= MAX_ACTIVE_REGIONS
) {
3411 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3412 MAX_ACTIVE_REGIONS
);
3416 early_node_map
[i
].nid
= nid
;
3417 early_node_map
[i
].start_pfn
= start_pfn
;
3418 early_node_map
[i
].end_pfn
= end_pfn
;
3419 nr_nodemap_entries
= i
+ 1;
3423 * shrink_active_range - Shrink an existing registered range of PFNs
3424 * @nid: The node id the range is on that should be shrunk
3425 * @old_end_pfn: The old end PFN of the range
3426 * @new_end_pfn: The new PFN of the range
3428 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3429 * The map is kept at the end physical page range that has already been
3430 * registered with add_active_range(). This function allows an arch to shrink
3431 * an existing registered range.
3433 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3434 unsigned long new_end_pfn
)
3438 /* Find the old active region end and shrink */
3439 for_each_active_range_index_in_nid(i
, nid
)
3440 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3441 early_node_map
[i
].end_pfn
= new_end_pfn
;
3447 * remove_all_active_ranges - Remove all currently registered regions
3449 * During discovery, it may be found that a table like SRAT is invalid
3450 * and an alternative discovery method must be used. This function removes
3451 * all currently registered regions.
3453 void __init
remove_all_active_ranges(void)
3455 memset(early_node_map
, 0, sizeof(early_node_map
));
3456 nr_nodemap_entries
= 0;
3457 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3458 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3459 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3460 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3463 /* Compare two active node_active_regions */
3464 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3466 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3467 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3469 /* Done this way to avoid overflows */
3470 if (arange
->start_pfn
> brange
->start_pfn
)
3472 if (arange
->start_pfn
< brange
->start_pfn
)
3478 /* sort the node_map by start_pfn */
3479 static void __init
sort_node_map(void)
3481 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3482 sizeof(struct node_active_region
),
3483 cmp_node_active_region
, NULL
);
3486 /* Find the lowest pfn for a node */
3487 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3490 unsigned long min_pfn
= ULONG_MAX
;
3492 /* Assuming a sorted map, the first range found has the starting pfn */
3493 for_each_active_range_index_in_nid(i
, nid
)
3494 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3496 if (min_pfn
== ULONG_MAX
) {
3498 "Could not find start_pfn for node %lu\n", nid
);
3506 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3508 * It returns the minimum PFN based on information provided via
3509 * add_active_range().
3511 unsigned long __init
find_min_pfn_with_active_regions(void)
3513 return find_min_pfn_for_node(MAX_NUMNODES
);
3517 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3519 * It returns the maximum PFN based on information provided via
3520 * add_active_range().
3522 unsigned long __init
find_max_pfn_with_active_regions(void)
3525 unsigned long max_pfn
= 0;
3527 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3528 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3534 * early_calculate_totalpages()
3535 * Sum pages in active regions for movable zone.
3536 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3538 unsigned long __init
early_calculate_totalpages(void)
3541 unsigned long totalpages
= 0;
3543 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3544 unsigned long pages
= early_node_map
[i
].end_pfn
-
3545 early_node_map
[i
].start_pfn
;
3546 totalpages
+= pages
;
3548 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3554 * Find the PFN the Movable zone begins in each node. Kernel memory
3555 * is spread evenly between nodes as long as the nodes have enough
3556 * memory. When they don't, some nodes will have more kernelcore than
3559 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3562 unsigned long usable_startpfn
;
3563 unsigned long kernelcore_node
, kernelcore_remaining
;
3564 unsigned long totalpages
= early_calculate_totalpages();
3565 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3568 * If movablecore was specified, calculate what size of
3569 * kernelcore that corresponds so that memory usable for
3570 * any allocation type is evenly spread. If both kernelcore
3571 * and movablecore are specified, then the value of kernelcore
3572 * will be used for required_kernelcore if it's greater than
3573 * what movablecore would have allowed.
3575 if (required_movablecore
) {
3576 unsigned long corepages
;
3579 * Round-up so that ZONE_MOVABLE is at least as large as what
3580 * was requested by the user
3582 required_movablecore
=
3583 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3584 corepages
= totalpages
- required_movablecore
;
3586 required_kernelcore
= max(required_kernelcore
, corepages
);
3589 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3590 if (!required_kernelcore
)
3593 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3594 find_usable_zone_for_movable();
3595 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3598 /* Spread kernelcore memory as evenly as possible throughout nodes */
3599 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3600 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3602 * Recalculate kernelcore_node if the division per node
3603 * now exceeds what is necessary to satisfy the requested
3604 * amount of memory for the kernel
3606 if (required_kernelcore
< kernelcore_node
)
3607 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3610 * As the map is walked, we track how much memory is usable
3611 * by the kernel using kernelcore_remaining. When it is
3612 * 0, the rest of the node is usable by ZONE_MOVABLE
3614 kernelcore_remaining
= kernelcore_node
;
3616 /* Go through each range of PFNs within this node */
3617 for_each_active_range_index_in_nid(i
, nid
) {
3618 unsigned long start_pfn
, end_pfn
;
3619 unsigned long size_pages
;
3621 start_pfn
= max(early_node_map
[i
].start_pfn
,
3622 zone_movable_pfn
[nid
]);
3623 end_pfn
= early_node_map
[i
].end_pfn
;
3624 if (start_pfn
>= end_pfn
)
3627 /* Account for what is only usable for kernelcore */
3628 if (start_pfn
< usable_startpfn
) {
3629 unsigned long kernel_pages
;
3630 kernel_pages
= min(end_pfn
, usable_startpfn
)
3633 kernelcore_remaining
-= min(kernel_pages
,
3634 kernelcore_remaining
);
3635 required_kernelcore
-= min(kernel_pages
,
3636 required_kernelcore
);
3638 /* Continue if range is now fully accounted */
3639 if (end_pfn
<= usable_startpfn
) {
3642 * Push zone_movable_pfn to the end so
3643 * that if we have to rebalance
3644 * kernelcore across nodes, we will
3645 * not double account here
3647 zone_movable_pfn
[nid
] = end_pfn
;
3650 start_pfn
= usable_startpfn
;
3654 * The usable PFN range for ZONE_MOVABLE is from
3655 * start_pfn->end_pfn. Calculate size_pages as the
3656 * number of pages used as kernelcore
3658 size_pages
= end_pfn
- start_pfn
;
3659 if (size_pages
> kernelcore_remaining
)
3660 size_pages
= kernelcore_remaining
;
3661 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3664 * Some kernelcore has been met, update counts and
3665 * break if the kernelcore for this node has been
3668 required_kernelcore
-= min(required_kernelcore
,
3670 kernelcore_remaining
-= size_pages
;
3671 if (!kernelcore_remaining
)
3677 * If there is still required_kernelcore, we do another pass with one
3678 * less node in the count. This will push zone_movable_pfn[nid] further
3679 * along on the nodes that still have memory until kernelcore is
3683 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3686 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3687 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3688 zone_movable_pfn
[nid
] =
3689 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3692 /* Any regular memory on that node ? */
3693 static void check_for_regular_memory(pg_data_t
*pgdat
)
3695 #ifdef CONFIG_HIGHMEM
3696 enum zone_type zone_type
;
3698 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3699 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3700 if (zone
->present_pages
)
3701 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3707 * free_area_init_nodes - Initialise all pg_data_t and zone data
3708 * @max_zone_pfn: an array of max PFNs for each zone
3710 * This will call free_area_init_node() for each active node in the system.
3711 * Using the page ranges provided by add_active_range(), the size of each
3712 * zone in each node and their holes is calculated. If the maximum PFN
3713 * between two adjacent zones match, it is assumed that the zone is empty.
3714 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3715 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3716 * starts where the previous one ended. For example, ZONE_DMA32 starts
3717 * at arch_max_dma_pfn.
3719 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3724 /* Sort early_node_map as initialisation assumes it is sorted */
3727 /* Record where the zone boundaries are */
3728 memset(arch_zone_lowest_possible_pfn
, 0,
3729 sizeof(arch_zone_lowest_possible_pfn
));
3730 memset(arch_zone_highest_possible_pfn
, 0,
3731 sizeof(arch_zone_highest_possible_pfn
));
3732 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3733 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3734 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3735 if (i
== ZONE_MOVABLE
)
3737 arch_zone_lowest_possible_pfn
[i
] =
3738 arch_zone_highest_possible_pfn
[i
-1];
3739 arch_zone_highest_possible_pfn
[i
] =
3740 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3742 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3743 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3745 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3746 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3747 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3749 /* Print out the zone ranges */
3750 printk("Zone PFN ranges:\n");
3751 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3752 if (i
== ZONE_MOVABLE
)
3754 printk(" %-8s %8lu -> %8lu\n",
3756 arch_zone_lowest_possible_pfn
[i
],
3757 arch_zone_highest_possible_pfn
[i
]);
3760 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3761 printk("Movable zone start PFN for each node\n");
3762 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3763 if (zone_movable_pfn
[i
])
3764 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3767 /* Print out the early_node_map[] */
3768 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3769 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3770 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3771 early_node_map
[i
].start_pfn
,
3772 early_node_map
[i
].end_pfn
);
3774 /* Initialise every node */
3775 setup_nr_node_ids();
3776 for_each_online_node(nid
) {
3777 pg_data_t
*pgdat
= NODE_DATA(nid
);
3778 free_area_init_node(nid
, pgdat
, NULL
,
3779 find_min_pfn_for_node(nid
), NULL
);
3781 /* Any memory on that node */
3782 if (pgdat
->node_present_pages
)
3783 node_set_state(nid
, N_HIGH_MEMORY
);
3784 check_for_regular_memory(pgdat
);
3788 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3790 unsigned long long coremem
;
3794 coremem
= memparse(p
, &p
);
3795 *core
= coremem
>> PAGE_SHIFT
;
3797 /* Paranoid check that UL is enough for the coremem value */
3798 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3804 * kernelcore=size sets the amount of memory for use for allocations that
3805 * cannot be reclaimed or migrated.
3807 static int __init
cmdline_parse_kernelcore(char *p
)
3809 return cmdline_parse_core(p
, &required_kernelcore
);
3813 * movablecore=size sets the amount of memory for use for allocations that
3814 * can be reclaimed or migrated.
3816 static int __init
cmdline_parse_movablecore(char *p
)
3818 return cmdline_parse_core(p
, &required_movablecore
);
3821 early_param("kernelcore", cmdline_parse_kernelcore
);
3822 early_param("movablecore", cmdline_parse_movablecore
);
3824 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3827 * set_dma_reserve - set the specified number of pages reserved in the first zone
3828 * @new_dma_reserve: The number of pages to mark reserved
3830 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3831 * In the DMA zone, a significant percentage may be consumed by kernel image
3832 * and other unfreeable allocations which can skew the watermarks badly. This
3833 * function may optionally be used to account for unfreeable pages in the
3834 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3835 * smaller per-cpu batchsize.
3837 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3839 dma_reserve
= new_dma_reserve
;
3842 #ifndef CONFIG_NEED_MULTIPLE_NODES
3843 static bootmem_data_t contig_bootmem_data
;
3844 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3846 EXPORT_SYMBOL(contig_page_data
);
3849 void __init
free_area_init(unsigned long *zones_size
)
3851 free_area_init_node(0, NODE_DATA(0), zones_size
,
3852 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3855 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3856 unsigned long action
, void *hcpu
)
3858 int cpu
= (unsigned long)hcpu
;
3860 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3861 local_irq_disable();
3863 vm_events_fold_cpu(cpu
);
3865 refresh_cpu_vm_stats(cpu
);
3870 void __init
page_alloc_init(void)
3872 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3876 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3877 * or min_free_kbytes changes.
3879 static void calculate_totalreserve_pages(void)
3881 struct pglist_data
*pgdat
;
3882 unsigned long reserve_pages
= 0;
3883 enum zone_type i
, j
;
3885 for_each_online_pgdat(pgdat
) {
3886 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3887 struct zone
*zone
= pgdat
->node_zones
+ i
;
3888 unsigned long max
= 0;
3890 /* Find valid and maximum lowmem_reserve in the zone */
3891 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3892 if (zone
->lowmem_reserve
[j
] > max
)
3893 max
= zone
->lowmem_reserve
[j
];
3896 /* we treat pages_high as reserved pages. */
3897 max
+= zone
->pages_high
;
3899 if (max
> zone
->present_pages
)
3900 max
= zone
->present_pages
;
3901 reserve_pages
+= max
;
3904 totalreserve_pages
= reserve_pages
;
3908 * setup_per_zone_lowmem_reserve - called whenever
3909 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
3910 * has a correct pages reserved value, so an adequate number of
3911 * pages are left in the zone after a successful __alloc_pages().
3913 static void setup_per_zone_lowmem_reserve(void)
3915 struct pglist_data
*pgdat
;
3916 enum zone_type j
, idx
;
3918 for_each_online_pgdat(pgdat
) {
3919 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3920 struct zone
*zone
= pgdat
->node_zones
+ j
;
3921 unsigned long present_pages
= zone
->present_pages
;
3923 zone
->lowmem_reserve
[j
] = 0;
3927 struct zone
*lower_zone
;
3931 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
3932 sysctl_lowmem_reserve_ratio
[idx
] = 1;
3934 lower_zone
= pgdat
->node_zones
+ idx
;
3935 lower_zone
->lowmem_reserve
[j
] = present_pages
/
3936 sysctl_lowmem_reserve_ratio
[idx
];
3937 present_pages
+= lower_zone
->present_pages
;
3942 /* update totalreserve_pages */
3943 calculate_totalreserve_pages();
3947 * setup_per_zone_pages_min - called when min_free_kbytes changes.
3949 * Ensures that the pages_{min,low,high} values for each zone are set correctly
3950 * with respect to min_free_kbytes.
3952 void setup_per_zone_pages_min(void)
3954 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
3955 unsigned long lowmem_pages
= 0;
3957 unsigned long flags
;
3959 /* Calculate total number of !ZONE_HIGHMEM pages */
3960 for_each_zone(zone
) {
3961 if (!is_highmem(zone
))
3962 lowmem_pages
+= zone
->present_pages
;
3965 for_each_zone(zone
) {
3968 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3969 tmp
= (u64
)pages_min
* zone
->present_pages
;
3970 do_div(tmp
, lowmem_pages
);
3971 if (is_highmem(zone
)) {
3973 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3974 * need highmem pages, so cap pages_min to a small
3977 * The (pages_high-pages_low) and (pages_low-pages_min)
3978 * deltas controls asynch page reclaim, and so should
3979 * not be capped for highmem.
3983 min_pages
= zone
->present_pages
/ 1024;
3984 if (min_pages
< SWAP_CLUSTER_MAX
)
3985 min_pages
= SWAP_CLUSTER_MAX
;
3986 if (min_pages
> 128)
3988 zone
->pages_min
= min_pages
;
3991 * If it's a lowmem zone, reserve a number of pages
3992 * proportionate to the zone's size.
3994 zone
->pages_min
= tmp
;
3997 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
3998 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
3999 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4002 /* update totalreserve_pages */
4003 calculate_totalreserve_pages();
4007 * Initialise min_free_kbytes.
4009 * For small machines we want it small (128k min). For large machines
4010 * we want it large (64MB max). But it is not linear, because network
4011 * bandwidth does not increase linearly with machine size. We use
4013 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4014 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4030 static int __init
init_per_zone_pages_min(void)
4032 unsigned long lowmem_kbytes
;
4034 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4036 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4037 if (min_free_kbytes
< 128)
4038 min_free_kbytes
= 128;
4039 if (min_free_kbytes
> 65536)
4040 min_free_kbytes
= 65536;
4041 setup_per_zone_pages_min();
4042 setup_per_zone_lowmem_reserve();
4045 module_init(init_per_zone_pages_min
)
4048 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4049 * that we can call two helper functions whenever min_free_kbytes
4052 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4053 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4055 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4057 setup_per_zone_pages_min();
4062 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4063 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4068 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4073 zone
->min_unmapped_pages
= (zone
->present_pages
*
4074 sysctl_min_unmapped_ratio
) / 100;
4078 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4079 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4084 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4089 zone
->min_slab_pages
= (zone
->present_pages
*
4090 sysctl_min_slab_ratio
) / 100;
4096 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4097 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4098 * whenever sysctl_lowmem_reserve_ratio changes.
4100 * The reserve ratio obviously has absolutely no relation with the
4101 * pages_min watermarks. The lowmem reserve ratio can only make sense
4102 * if in function of the boot time zone sizes.
4104 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4105 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4107 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4108 setup_per_zone_lowmem_reserve();
4113 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4114 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4115 * can have before it gets flushed back to buddy allocator.
4118 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4119 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4125 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4126 if (!write
|| (ret
== -EINVAL
))
4128 for_each_zone(zone
) {
4129 for_each_online_cpu(cpu
) {
4131 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4132 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4138 int hashdist
= HASHDIST_DEFAULT
;
4141 static int __init
set_hashdist(char *str
)
4145 hashdist
= simple_strtoul(str
, &str
, 0);
4148 __setup("hashdist=", set_hashdist
);
4152 * allocate a large system hash table from bootmem
4153 * - it is assumed that the hash table must contain an exact power-of-2
4154 * quantity of entries
4155 * - limit is the number of hash buckets, not the total allocation size
4157 void *__init
alloc_large_system_hash(const char *tablename
,
4158 unsigned long bucketsize
,
4159 unsigned long numentries
,
4162 unsigned int *_hash_shift
,
4163 unsigned int *_hash_mask
,
4164 unsigned long limit
)
4166 unsigned long long max
= limit
;
4167 unsigned long log2qty
, size
;
4170 /* allow the kernel cmdline to have a say */
4172 /* round applicable memory size up to nearest megabyte */
4173 numentries
= nr_kernel_pages
;
4174 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4175 numentries
>>= 20 - PAGE_SHIFT
;
4176 numentries
<<= 20 - PAGE_SHIFT
;
4178 /* limit to 1 bucket per 2^scale bytes of low memory */
4179 if (scale
> PAGE_SHIFT
)
4180 numentries
>>= (scale
- PAGE_SHIFT
);
4182 numentries
<<= (PAGE_SHIFT
- scale
);
4184 /* Make sure we've got at least a 0-order allocation.. */
4185 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4186 numentries
= PAGE_SIZE
/ bucketsize
;
4188 numentries
= roundup_pow_of_two(numentries
);
4190 /* limit allocation size to 1/16 total memory by default */
4192 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4193 do_div(max
, bucketsize
);
4196 if (numentries
> max
)
4199 log2qty
= ilog2(numentries
);
4202 size
= bucketsize
<< log2qty
;
4203 if (flags
& HASH_EARLY
)
4204 table
= alloc_bootmem(size
);
4206 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4208 unsigned long order
;
4209 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4211 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4213 * If bucketsize is not a power-of-two, we may free
4214 * some pages at the end of hash table.
4217 unsigned long alloc_end
= (unsigned long)table
+
4218 (PAGE_SIZE
<< order
);
4219 unsigned long used
= (unsigned long)table
+
4221 split_page(virt_to_page(table
), order
);
4222 while (used
< alloc_end
) {
4228 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4231 panic("Failed to allocate %s hash table\n", tablename
);
4233 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4236 ilog2(size
) - PAGE_SHIFT
,
4240 *_hash_shift
= log2qty
;
4242 *_hash_mask
= (1 << log2qty
) - 1;
4247 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4248 struct page
*pfn_to_page(unsigned long pfn
)
4250 return __pfn_to_page(pfn
);
4252 unsigned long page_to_pfn(struct page
*page
)
4254 return __page_to_pfn(page
);
4256 EXPORT_SYMBOL(pfn_to_page
);
4257 EXPORT_SYMBOL(page_to_pfn
);
4258 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4260 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4261 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4264 #ifdef CONFIG_SPARSEMEM
4265 return __pfn_to_section(pfn
)->pageblock_flags
;
4267 return zone
->pageblock_flags
;
4268 #endif /* CONFIG_SPARSEMEM */
4271 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4273 #ifdef CONFIG_SPARSEMEM
4274 pfn
&= (PAGES_PER_SECTION
-1);
4275 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4277 pfn
= pfn
- zone
->zone_start_pfn
;
4278 return (pfn
>> (MAX_ORDER
-1)) * NR_PAGEBLOCK_BITS
;
4279 #endif /* CONFIG_SPARSEMEM */
4283 * get_pageblock_flags_group - Return the requested group of flags for the MAX_ORDER_NR_PAGES block of pages
4284 * @page: The page within the block of interest
4285 * @start_bitidx: The first bit of interest to retrieve
4286 * @end_bitidx: The last bit of interest
4287 * returns pageblock_bits flags
4289 unsigned long get_pageblock_flags_group(struct page
*page
,
4290 int start_bitidx
, int end_bitidx
)
4293 unsigned long *bitmap
;
4294 unsigned long pfn
, bitidx
;
4295 unsigned long flags
= 0;
4296 unsigned long value
= 1;
4298 zone
= page_zone(page
);
4299 pfn
= page_to_pfn(page
);
4300 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4301 bitidx
= pfn_to_bitidx(zone
, pfn
);
4303 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4304 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4311 * set_pageblock_flags_group - Set the requested group of flags for a MAX_ORDER_NR_PAGES block of pages
4312 * @page: The page within the block of interest
4313 * @start_bitidx: The first bit of interest
4314 * @end_bitidx: The last bit of interest
4315 * @flags: The flags to set
4317 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4318 int start_bitidx
, int end_bitidx
)
4321 unsigned long *bitmap
;
4322 unsigned long pfn
, bitidx
;
4323 unsigned long value
= 1;
4325 zone
= page_zone(page
);
4326 pfn
= page_to_pfn(page
);
4327 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4328 bitidx
= pfn_to_bitidx(zone
, pfn
);
4330 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4332 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4334 __clear_bit(bitidx
+ start_bitidx
, bitmap
);